Archived

Graphcore was acquired by SoftBank — this page is preserved as an archive. Visit fobi.com for current frontier-tech coverage.

Graphcore

Acquired by SoftBank · 2024

Graphcore designs Intelligence Processing Units (IPUs) for AI training and inference, targeting machine intelligence workloads. Raised $700M+ but struggled to compete with Nvidia; layoffs in 2023. Bow IPU launched but adoption limited. Acquired by SoftBank in July 2024; SoftBank injected a further $457M in May 2026.

Graphcore was an early AI chip challenger with significant funding and IPU architecture, though momentum has slowed vs GPU rivals.

graphcore.ai AI AcceleratorsUKacquired

At a glance

Total raised

$532M

last round acquired

Headcount

—

no public estimate on record

Headquarters

Bristol

Founded

2016

10 years active

Status

Acquired

by SoftBank

Patents on file

USPTO grants

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

91 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$532M

Rounds tracked

Latest stage

Acquired

Date	Stage	Amount	Valuation	Lead investors
Jul 11, 2024	Acquired	$500M	—	SoftBank
Dec 29, 2020	Series E	$222M	$2.8B	Ontario Teachers' Pension Plan Board
Dec 1, 2018	Series D	$200M	$1.7B	Microsoft, Samsung, Dell Technologies
Nov 1, 2017	Series C	$50M	—	Sequoia Capital
Jul 1, 2017	Series B	$30M	—	Atomico
Oct 1, 2016	Series A	$30M	—	Robert Bosch Venture Capital

How they make money

Hardware (one-time)Pattern

Graphcore sells a physical product — the Intelligence Processing Unit (IPU) — as a one-time hardware sale to enterprises and cloud providers building AI infrastructure. Customers purchase the Bow IPU as a discrete accelerator card or as part of an IPU-POD system for training and inference workloads. Revenue is recognized upon delivery of the hardware, with no disclosed recurring software or service component generating material ongoing revenue. The economics are those of a fabless semiconductor company: Graphcore designs the chip and software stack, while third-party foundries handle manufacturing. Gross margin depends on wafer costs, package yields, and selling price — all closely tied to performance-per-dollar versus Nvidia's GPUs. Volume is the primary scaling lever, but adoption has been limited, making it difficult to amortize the high fixed costs of chip design across a large installed base. The software stack (Poplar) is provided to enable the hardware but does not itself carry a separate license fee in the public model.

Generated by deepseek-v4-flash · June 2, 2026Analytical summary · not investment advice

Founder backgrounds

Nigel Toon

Co-founder & CEO

A British semiconductor entrepreneur with a career spanning multiple successful chip company exits. Began at Altera Corporation, where he served as VP and GM for Europe. Co-founded Icera, a 3G cellular modem chip company, in 2002; Nvidia acquired Icera in 2011 for $367 million. Also served as CEO of XMOS and Picochip (acquired by MiMedia, now Intel). Co-founded Graphcore in Bristol in 2016 alongside Simon Knowles after they met at a pub in Bath to identify the next major opportunity in processor design — specifically a new architecture purpose-built for machine intelligence. Toon has led the company through multiple funding rounds totaling over $700 million, including SoftBank's 2023 acquisition.

Simon Knowles

Co-founder & former CTO (exited Aug 2025)

Holds a degree in electrical engineering from the University of Cambridge and more than 20 patents in microprocessor design. Co-founded Element14, a processor IP company acquired by Broadcom in 2000 for approximately $600 million. Subsequently co-founded Icera alongside Nigel Toon; Nvidia acquired Icera in 2011 for $367 million. The two reunited to found Graphcore in Bristol in 2016, where Knowles served as CTO and EVP of Engineering, architecting the Intelligence Processing Unit (IPU) — a massively parallel processor designed from the ground up for machine learning graph computation. He was elected a Fellow of the Royal Society and exited Graphcore in August 2025 following SoftBank's acquisition.

AI bull / bear

Bull case

SoftBank's backing gives Graphcore the war chest and strategic patience to outlast the current GPU-dominated cycle. With over $700M raised, the company has the runway to keep refining its IPU architecture for inference workloads, where Nvidia's advantage is narrower than in training. The Bow IPU's novel architecture was purpose-built for sparse compute and graph-based models, a class of workloads that will grow as recommendation systems, molecular simulation, and graph neural networks gain commercial traction. If inference efficiency overtakes raw training throughput as the industry's binding constraint within the next 18-24 months, Graphcore's differentiated design could become the default choice for specialized deployment. SoftBank also opens distribution channels into its massive portfolio of portfolio companies, giving Graphcorp a captive addressable market it never had as an independent startup.

Bear case

Graphcore raised $700M+ and still could not break Nvidia's grip on AI compute. The 2023 layoffs and limited Bow IPU adoption signal the it has already lost the market's confidence. Nvidia's CUDA moat is not just a software advantage; it is a network effect that makes any alternative device a second-class citizen for developers. Enterprises will not bet on a niche architecture when the entire AI software stack is built around CUDA and Triton. Meanwhile, well-funded rivals like Cerebras, Groq, and AMD are also chasing the same inference opportunity with better commercial traction. For Graphcore, the window to achieve product-market fit has essentially closed. SoftBank's ownership may provide a soft landing or an acqui-hire, but it does not solve the core problem: the IPU does not run the models customers actually use, and developers have no incentive to port their code to it.

Generated by deepseek-v4-flash · June 2, 2026Refreshes quarterly · see disclosures

In the news

Graphcore · last 30 days

No articles ingested yet for Graphcore. Once the hourly news pipeline is live, every article the classifier tags as mentioning this company appears here with its one-line AI summary and sentiment.

Watch

Leadership

Nigel Toon

Co-founder & CEO

—

Simon Knowles

Co-founder & former CTO (exited Aug 2025)

—

Products

Bow IPU (IPU-POD)
Graphcore's third-generation Intelligence Processing Unit, the first processor to use wafer-on-wafer 3D stacking (with TSMC) to boost power delivery and performance. The Bow IPU packs ~1,472 independent cores with large on-chip SRAM, designed for the fine-grained parallelism of machine-intelligence workloads, and is deployed in scale-out IPU-POD systems. It is the flagship of Graphcore's shipping hardware line, though adoption has trailed Nvidia GPUs.
Poplar SDK
Graphcore's software stack—compiler, graph engine, and framework integrations for PyTorch and TensorFlow—that maps AI models onto the IPU's many cores and explicit on-chip memory. Poplar is essential to programming the unconventional IPU architecture and exposing its parallelism to developers. As the bridge between standard ML frameworks and Graphcore silicon, it is core to the company's platform and any future products.
Izanagi (next-generation, in development)
Graphcore's next-generation AI compute platform under development since the 2024 SoftBank acquisition, reported to target large-scale deployment across 'Stargate' hyperscale data centers in the US and Japan in 2026. Positioned for frontier-scale AI training and inference under SoftBank's broader AI-infrastructure strategy, Izanagi represents Graphcore's reset toward hyperscale systems rather than competing chip-by-chip with merchant GPUs.

Most recent patents

41 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.